On 25/11/2024 17:41, Guenter Roeck wrote: >>>>> >>>>> with the assumption that Lee will pick it up. >>>> >>>> This was merged, while I was AFK, so the ship has sailed, but let me >>>> state here objection for any future discussions: >>>> >>>> NAK, this is not a thermal sensor. The commit msg explains what they >>>> want to achieve, but that's not a valid reason to add property from >>>> different class of devices. >>>> >>>> This is some hardware/temperature monitoring device or power supply, not >>>> part of SoC, not integrated into any SoC thermal zone. Calling it >>> >>> I am confused. We have several thermal sensors registering as thermal >>> zone, and fan controllers registering themselves as thermal cooling devices. >>> >>> Are you saying that this is all not permitted because they are not part >>> of a SoC ? >> >> >> These are fine, because they monitor or cool down the SoC. Sensor can >> be under the die. Fan for battery or for battery charger also would be >> fine, because it is a real cooling device. It literally cools. >> > > Sorry, I don't get the distinction since you specifically refer to the SoC. > How about drive temperatures ? RAM temperatures ? Temperatures reported Several of them are part of the SoC, like DDR. But even if they are not, I agree they could be a thermal sensor, but I would stop before calling them thermal zones and this patchset leads to such calling. > by networking chips ? Power supply temperatures ? We have all those registering > thermal zones. The ASUS EC controller driver registers thermal zones > for its temperature sensors. Dell and HP drivers do the same. Maybe we need to clarify that thermal sensors and zones are not specific to SoCs? For now all bindings say: "thermal-sensor: device that measures temperature, has SoC-specific bindings" > >> But treating battery charger as cooling device is not correct, IMHO. > > Confused. The patch you object to isn't introducing a cooling device, > it is introducing #thermal-sensor-cells. The EC reports temperature The next patchset is. This is one of the problems with series from Sung-Chi, Li - they add hardware description piece by piece, to match the driver needs, while we expect to see complete hardware picture. In the complete picture (f:lschyi@xxxxxxxxxxxx in lore) you would see the battery being called a cooling sensor with explanation: "The cros_ec supports limiting the input current to act as a passive thermal cooling device. Add the property '#cooling-cells' bindings, such that thermal framework can recognize cros_ec as a valid thermal cooling device." > sensor results, and the driver wants to register those as thermal zones. > Yes, it may well be that one of those temperature sensors is close to > a battery charger, but I am not even sure if that is really the case. Hm, my impression based on very limited commit msg was that it is about battery. Probably I don't understand how the hardware looks here, but sorry, commit msgs and bindings descriptions are for a reason - to help me understanding it. > >> Battery charger does not cool anything down and already we have there >> properties for managing thermal and current aspects. >> > Agreed, but unless I am missing something that isn't done here. Yep, I connected two separate patchsets, because they form greater work of making power supply a cooling device, AFAIU. > >> BTW, if power supply bindings miss some thermal aspects, then let's grow >> the common binding first and agree on common aspects. >> > > I don't even know how the two patches are associated with power supplies > or battery chargers in the first place. All they try to do is to enable > adding the temperature sensor values reported by the EC in Chromebooks > to thermal zones. I don't recall any previous limitations on the ability > to register thermal sensors as thermal zone with the thermal subsystem, > and I am trying to understand what those limitations are. > > So far my approach was "ok, someone wants to register a thermal sensor as > thermal zone - fine, let's do that. We have close to 50 thermal sensors on > a variety of devices - including but not limited to disk drives, memory, > Ethernet controllers, Ethernet PHYs, SFPs, RTCs, and ECs - registering > as thermal zones from hardware monitoring drivers. I don't recall anyone > ever saying "no, you can't do that". > > I am trying to understand if some of those are inappropriate and, if so, > why that is the case. Probably would be nice to drop remaining references to SoC from thermal bindings and just interpret thermal zones as system-wide. I retract than my objects against sensor cells, but I keep my objection of changing these bindings piece by piece. This should be one complete work for bindings so we see what this hardware is supposed to do. Best regards, Krzysztof
